This thesis, titled "Quantum-Aware Image Encoding and Adversarial Perturbation," explores a hybrid system designed to protect digital creators from unauthorized generative AI training. It bridges classical adversarial defense with Quantum Image Processing (FRQI/QPIXL) to create "cloaks" that remain effective even after quantum encoding.
Key Findings:
• High Robustness: 99.99% of classical adversarial perturbations survive the quantum encoding process, proving the reliability of hybrid pipelines.
• Invisible Protection: The system maintains high visual fidelity with a mean PSNR of 32.30 dB and an SSIM near 1.0, ensuring protections are imperceptible to humans.
• NISQ-Ready: The framework uses a modular, patch-based architecture specifically optimized for current Noisy Intermediate-Scale Quantum (NISQ) hardware.
• Effective Defense: Successfully misleads advanced feature extractors like DINOv2, providing a scalable foundation for digital sovereignty against future quantum-accelerated AI.
(write your relevancy here)
(write your future scope here)
(write your Skills acquired here)
